The paper describes quite sophisticated modelling system to simulate CO2 emission and dispersion with a  hectometer resolution eddy-resolving model DALES.                                                  

The work presents a very interesting experience of using LES for to simulate CO2, an is definitely of interest for ACPD readership. Several issues remain to be clarified before the paper can be accepted.

General points

The paper gets into too much very technical details about implementations, while some important characteristics of the simulations are missing.  With a minor effort one can calculate DALES domain is of about 1100x1500 cells.  How many layers are there and what are they? How the boundary conditions for mean wind, for turbulence and for tracers have been implemented both in lateral and in vertical?  How much spin-up does  the simulation needs to become realistic? etc.

I could not find how surface emissions are handled. Are they just injected into the lowest LES layer? 

There is little information on the reasons why specific technical decisions were taken. It is not clear why one would need to drive an ecosystem model for CO2 fluxes with high-resolution LES fields?  Are the features such as second-scale updates of meteorology, clouds, physics of the ABL really needed to simulate CO2 fluxes by ecosystem, or are they just an overhead for such simulations? Same applies to plume-rise calculation.

There is not much info on the computational costs of the DALES CO2 simulations.  How many cores and hours needed to simulate one hour of the dispersion?  How well it scales with number of cores?  How much is the overhead for CO2 with respect to plain DALES?  What are cost-benefits of using LES vs using e.g. LOTOS-EUROS at similar resolution for practical applications?

Some of the used datasets have been declared public, but the procedure of obtaining them has not been clearly described.  "The traffic data shapefile is provided by the  Dat.mobility company and can be requested from RIVM." Publishing these assets via some open data publishing service or clearly identifying the specific dataset and providing a contact or URL would make it more clear.

The text is full of vague constructs that do not bring much information: "high-accuracy", "realistic approach", "fine resolution" etc.  In a scientific paper each of such constructs if used has to have a very specific meaning.

Specific comments:

Table 1 leaves a question on the quality of the fits used. It could be better replaced with regression plots used for get the fit. Then one could judge how good or bad the used approximation is. A simple three-panel figure would do: (a) log(flow rate) vs emission rate (of which species?); (b) Exhaust temperature vs emission rate for specific processes; (c) log height vs  emission rate.

Fig1 It might make sense to mention that SNAP8 includes shipping. SNAP10 has some marine emissions.  What is the marine agriculture ?

Fig1 and Fig2 are labeled to have emissions in kg/year. Normally I would guess that it  was meant kg/year/cell, however

the figure has at least two distinct cell sizes, which is confusing. Should it be kg/year/ha then? It might make sense to mark the LES domain on these maps, to indicate that missing emissions from Germany (SE corner of the maps) do not affect the simulations.

l215. The shape file has not been specified properly.

l220 What is the advantage of downscaling 1km traffic emissions instead of just assigning CO2/NOx emission factors per vehicle type and use NOx inventory? At least the latter approach would give more consistent map.

l260 What is the benefit of driving A-gs model with a few-second resolution? Is there any benefit in having it online with turbulence-resolving model?

 l287 Is the model really able to conduct simulations and examine specific aspects of anything? I would say it should be a modeler..

l290-310: Sec 3.1 describes unit conversion and linear interpolation in time. Both are quite trivial and,

probably not worth the paper space.  Same applies to appendix B.

Sec 3.2 describes quite standard treatment of plume rise. A reference to Gordon (2018) and Akingunola (2018) would be sufficient to describe that, unless something different has been implemented in DALES. In the latter case a brief description with  highlighting the differences would be needed.

Table A1 should have a proper reference.

The manuscript entitled: "Carbon dioxide plume dispersion simulated at hectometer scale using DALES: model formulation and observational evaluation" by Karagodin-Doyennel et al., is a comprehensive study in which CO₂ mole fractions for a domain in The Netherlands (NL) were simulated using DALES, an LES model that runs at 100x100 m. Anthropogenic emissions (traffic, power, agriculture, residential, among others) and biogenic fluxes are added to the LES framework, so all flux components contributing to total CO₂ mole fractions are taken care of. A specific processing pipeline is described for the downscaling of anthropogenic emissions to achieve the desired spatial scale. Finally, the CAMS air quality forecast is used as the CO₂ background for the LES domain.  The evaluation is done at 3 sites in the NL: Cabauw, Slufter and Westmaas using simulated tagged tracers to investigate the main drivers of variability at each site. There is quite some work on this study, the Figures are very well produced and overall the paper is well-written. However, the following are major points for improvement that the authors should consider before the article is accepted for publication:

General Comments

1. The article is presented with a lot of focus on anthropogenic emissions, which makes sense given the motivation in the Introduction hinting at the Paris agreement, emission reductions, etc. Therefore, the effort put on Section 2,  preparing the SNAP categories, adjusting the point and diffuse area sources, downscaling emissions, plume rise, etc., is justified and well explained. However, in the Results from Figure 7 to 10, I was surprised that the authors do not have a single tracer with Anthropogenic emissions or even multiple tracers for each SNAP category. Instead, they show CO₂bg, CO₂sum, and CO₂NEE with no opportunity to evaluate the individual contribution of the anthropogenic emissions. This is somewhat contradicting the set and setting given in the Introduction and all the effort done processing the emissions. Why not showing at least the CO₂bg_emiss?

2. The use of CAMS as background is one of my major concerns. First, it is not entirely clear which CAMS product is used, the link provided ends up in the ADS website and from there several products providing CO₂ mole fractions can be used, for example: CAMS global greenhouse gas forecasts, CAMS global greenhouse gas reanalysis (EGG4) or CAMS global inversion-optimised greenhouse gas fluxes and concentrations. Second,  as I suspect the forecast was used, for the modelling exercise the authors are performing they should show that the background does a good job in background stations and for this study in continental stations. From my experience at other latitudes, the CAMS forecast is biased low and likely would have an offset for your domain as well. I suggest to investigate this further and if needed bias correct the CAMS background used. The situation that the authors are simulating, with northeasterly winds will bring signals from the Groningen Gas Field, likely bringing CO₂  signals that will be present in the observations but not in CAMS (I think). You can, for example, compare CAMS CO₂ at Cabauw and at a background stations, like Mace Head (Ireland) and from there calculate a bias, if any, and correct your CO2bg tracer. But this analysis, leading or not to a bias correction, should be shown in Supplementary material.

3.  Length of article and structure. As I said before, there is a lot of work done here, but the reader probably does not want to read all about it, so I suggest to shorten the article and change the structure. In the Methods, I suggest to start with the DALES description and then continue with the emission preprocessing. In the Results, I strongly suggest to show the evaluation at the three sites all together, or start with Slufter and Westmass and then Cabauw. Now, the authors start by showing the weak part of the study: the comparison at Cabauw where LOTOS-EUROS performs better in most of the analysed statistcs. In addition, in each individual section of the Results the description of the plots are very subjective, using statements like: "typical diurnal variability", "slightly overestimate", "well-pronounced", "shows the tendency", "degree of correspondence", "lower accuracy", "levels that might be difficult to reproduce", etc. If you use the numbers you have at hand, this can be shortened significantly. One more thing, you devote quite some text to plume rise, but this is not discussed nor evaluated in the Results, consider shortening the plume rise part considerably or giving it more protagonism in your results. 

4. Given that the authors include biogenic fluxes and that in the Results this signal is important for the interpretation, I think they should: i) explain with more detail how are biogenic fluxes simulated or given as offline input to LOTOS-EUROS, ii) present the spatial distribution of grasslands and forests in DALES iii) if possible and the models permit it, evaluate the DALES/LOTOS-EUROS NEE fluxes at a flux tower within the domain.

5. As a GMD paper and given that one of the objectives of the paper is to "Document" the downscaling of the emission inventory, I think the authors should provide clear links to function names and modules in the code and give an overall idea of how the preprocessor was constructed. For example in line 141, the authors write: "The workflow is structured as a comprehensive program with several stand-alone modules, each responsible for different aspects of emission data processing". So I suggest you present an overview of the program and those modules that you are referring to. 

Specific comments:

L124: define E-PRTR, not done previously

L134:  the 100m spatial res. is for tracer transport but not for resolving turbulence? not clear since model is not described first.

L148-149: "1316 out of 1914" add a percentage here.

L156: I suggest to merge this two lines with the previous paragraph. In addition, what do you mean with "cohesive"?

L165: "In essence" may be "originally" fits better?

L167: Can you show an example of this separation on a Map? It will be informative to show the reader what is the output of this. Sufficient as SupFigure.

L169: "gpkg" is this a specific format in a given programming language or platform, specify.

L175: What about mass conservation in this procedure? Did you check that?

L176: This line reads like a caption.. I suggest removing it and start with the next paragraph.I dont think you need this sentence here.

Figure 1. Can you point the reader to what do you consider as point sources and area emissions on the Figure? For example, for SNAP 2: Res. and Commercial. Are these considered point sources or area emissions? SNAP 5 is quite vague: Fossil Fuels? most of these can fall in SNAP 5. Given the level of detail you are aiming, I suggest to be more precise here.  These SNAP categories represent individual tracers in DALES? Not clear.

L187: "responsible for the majority" how much, add %.

L188-190: This is not clear. Are there less sources in the northeast? Why only for evaluation of the model? Are you only simulating when the wind comes from the NE? Please clarify.

Figure 2.  I suggest to have Fig 1 and Fig 2 in one panel together. You can have Fig 2 as a large panel above the small panels of Fig 1. This will improve the visualization. "These emissions maps" should be singular instead?

L194: So, in the model the plume rise mechanism is always activated, but for area emissions the bottom and top are set to fixed values, right?What is the interaction between plume rise and “natural buoyancy”? As I think of plume rise in emissions, their rising is due to temp gradients of the emissions and surrounding air. But this is not the case for area emissions… clarify please.

L196-197: This contradicts the sentence in line 193 and 194. I suggest you describe the plume rise for point sources first and then for area emissions. It is confusing as it is now.

L200: The temporal disaggregation is for all categories? only for traffic? do all have a diurnal cycle or rush hour? I dont think so. How do you deal with Agriculture and Industrial processes for example?

L202-203:  you refer several times to the model description on subsequent section. I suggest you describe DALES first and then its inputs (emissions).

L210-211: "Data have 100 × 100 m2 resolution are freely available from the CBS website" something wrong in this sentence: "Data with 100 x 100 m2 resolution are freely...."

L222: "target level" add "target level (100m)"

L225-230: this sounds more like Discussion.

L231-234. With Figure 3 I am convinced, no need to state this in the Methods. In addition, so far I have not seen any reference to the downscaling code or repository.

L255: What about autotrophic respiration

L271-272: Do you also need a high resolution map of the forests in NL? Im thinking about the big parks within cities and off course the Veluwe.

L286: van Genuchten? what is that?

L294: which workflow? the previously described emission downscaling pipeline? Be specific here.

L301-302: How many vertical layers does the model have?

L310: you meant point sources?

L316: this line is not needed I think.

L331:  why not just say negative…?

L341: u_zhj+1 and uzhj are the wind speed using the three components of the wind ? u, v and w?

L394-395: Well. I don't understand why you took a lot of effort defining point sources, area sources, downscale, regrid, etc, to have all as one tracer in DALES. I really was expecting to see tracers per SNAP category. This makes me think, why all the effort of Sections 2.1 and 2.2 to group all anthropogenic tracers together! Now the way you are setting up the experiments suggest you are more interested in the biogenic signals.

L404-405: Why? this is also mentioned before. Be specific why this is the case. Not clear.

L406: In other words, you do like a 1-day spinup?

L420: I suggest merging this paragraph with the one above

L424: merge with previous paragraph… same topic..

Section 5.2 how are biogenic emissions treated in LOTOS-EURO? The background is also CAMS?

Figure 6.  add the names of the stations somewhere on the Figure.

L483: The CO2bg was not really computed by DALES right? Just taken from CAMS and transported in DALES?

L484-485: Using anomalies instead of full simulated mole fractions. So, why do you include a background? Why not full mole fractions and leave the observations untouched? You could also just compare enhancements.. Not clear why this is selected.

L486-487: first time NEE comes up. I assume this is the difference between soil resp and gpp in DALES. Not clear how is it for LOTOS-EUROS.

L488-489: Why you do a spatial operation to solve for temporal dynamics? Not clear. Maybe Im not getting what you are doing. I think what you want to say is: "DALES is sampled at the Cabauw tower interpolating horizontally and vertically to match the Cabauw measured CO₂ profile." correct?

L490-495: I would expect some statistics like MBE, RMSE, R2, instead of this subjective description. 

L501-503: see my comment above about the CAMS selection as background. 

L509-511: biases, you can quantify this, I suggest doing it. Overestimation by how much? you have numbers, use them.

L513: what is this? vertical component? if this is not to blame why do you mention it? Confusing.

L515-518: Sounds like methods. Also, you are bringing the statistics quite late in the Results. 

L519: Bootstrap of MAE? Not clear why this is done. Not mentioned in the Methods.

L521-522: It is not surprising that CO2sum is better than CO2bg.. I would not even report this. It is expected.

L525-526: its quite surprising that the mesoscale model explains 12% more of the variability than DALES. Why then going to LES scales ? (I know you show this later, but during the first reading I was wondering about this because the good DALES performance is presented after this).

L520-535:  most of the statistics show LOTOS-EUROS is better than DALES.. why then using the LES? (Again this to show what the reader thinks when this is presented first). Additionally, I think you should fix the background and make the comparison using total mole fractions and not anomalies. Or explain clearly why the selection of the anomalies for this analysis.

Figure 8. Why this plot shows full mole fractions? I suggest to be consistent either with anomalies like in Fig 7 or total mole fractions like this one.

L539: you meant: ”higher performance”? maybe change “higher” for “better”?

L541-542: what if you compare gradients in both models? Maybe DALES is better than LOTOS here?

L546-549: this sounds like methods, Maybe you can add a subSection in Methods called: "Model evaluation at atmospheric stations" and there tell the reader how do you sample DALES for each station and the height used.

Figure 9. Here it would be very interesting to have one single tracer for only anthropogenic emissions. You have one for NEE and not for the SNAP categories for which you have done most of the efforts  processing data. With this you can easily argument that these stations are more affected by anthropogenic sources than Cabauw. The three tracers you have: *sum, *NEE and *bg don't allow to make such analysis.

L563: "Anthropogenic emissions dominate in the CO2 variability in this location", you could show this clearly if you would had an anthropogenic emission tracer. Now, is difficult to see if the model gets that.

L551-576: A lot of subjective statements: slightly, typical, well-pronounce, etc. Only once numbers are used (L560).

L578-579: Again, this (CO2sum better than CO2bg) is not surprising. I would not even report this.

L587-592: Not clear what the message is in this paragraph. First you refer to both models being off with low R2, then you refer to the normalized std being better in DALES. Then you go back to both models being off with a high RMSD. Think about a message and argument it. Now this paragraph is difficult to follow.

Section 6.3: After reading this section, there was no clear message in the end. I had to go back and re-read to end up with no conclusive statement that summarizes the pros and cons at Wetsmaas and Slufter. You could refer to the difficulties the models have at Slufter, as you do now, but also make a conclusive statement for Westmaas. DALES was better there, so you can highlight this at the end of the Section.

L598-600: I had to wait 26 pages to see this. I think this should be presented earlier. Not sure if it is because the paper is too long, that it certainly feels quite late. I was wondering about the Anthropogenic emissions and their contribution from the beginning of the Results.

Figure 11. Nice plot. Some things to clarify: 

- the % is taken from the full mole fraction or from dCO2?

- not clear if dCO2 is from the model or from the obs?

- if dCO2 is based on the model, then consider adding information on the dCO2 from the observations. You could do this by subtracting the background from the observations, making sure you have bias corrected CAMS if needed.  See my comments above about this.

- I dont understand why you first use anomalies, then for the scatter plots you use full mole fractions and now, deviations from background Why not anomalies here? Why dCO2 since the beginning? Please argument this.

L628-629: This is true. But for making this statement you should show you are really doing this. I did not see a map (100x100m) of vegetation for your domain, it is not clear if you considered autotrophic respiration in DALES and there is no validation of NEE from DALES with eddy fluxes somewhere in NL. As you presented the Introduction and the Methods, the reader assumes the focus is on anthropogenic emissions and in the results there is a shift in focus to NEE.

Section 7. What about the perspective of LES for atmospheric inversions? something to mention here? You can speculate on optimizing specific SNAP categories and use this as a benchmark for what the country or city is reporting (bottom-up).

L645-649: I doubt higher resolution will be the main solution to the CAMS background. To get continental CO₂ signals properly you need at least to optimize the fluxes (inversion) and have a good atmospheric transport. The CAMS version you are using doesn't do that, I believe so.. not clear either. 

L689: access or link to code for downscaling, not seen so far.

L704. "also identified limitations of the current framework" like?

L717: "with inversion techniques"  This should be mentioned earlier in the Perspective section. How and what is the feasibility and also what type of information can it bring.

L718: "sub-scale processes" you meant sub-grid? also give examples.

L720-721: "will support more informed decision-making" How? I don't see this for the near future with a couple of days of simulations. I expect to see discussion on what are the perspectives of making longer simulations for at least 6 months so that they are relevant to policy.

Code and data availability: the offline workflow repo should be referenced in the main text. The DALES 4.4 repo should also be reference earlier.

Table A1. I suggest you add this on the main text, together with a description if they are considered point or area sources.

With this my opinion is to reconsidered the manuscript after major revisions.